Affiliations: [a]
School of Information Engineering, Wuyi University, Jiangmen, China
| [b]
School of Information Science and Technology, Zhongkai University of Agriculture and Engineering, Guangzhou, China
| [c]
Center for Educational Technology, Guangdong University of Finance and Economics, Guangzhou, China
| [d]
College of Science, Guilin University of Technology, Guilin, China
| [e]
Department of Mathematics and Computer Science, California State University, East Bay, USA
Correspondence:
[*]
Corresponding author. Ken Cai, School of Information Science and Technology, Zhongkai University of Agriculture and Engineering, Guangzhou 510225, China. Tel./Fax: +8602089002069; E-mail: icken@126.com.
Abstract: Diagnosis of rice planthopper pests based on imaging technology is an efficient means to develop intelligent agriculture. Effective contour automation extraction is an important pretreatment technology at the early stage for identifying and classifying rice planthoppers. The traditional graph cut method-based active contour (GCBAC) requires human–computer interaction during segmentation. In addition, GCBAC is prone to shrinking bias phenomenon, thereby providing short-boundary segmentation results. This study proposed a novel approach to overcome these two problems. First, rice planthopper initial segmentation was completed through discrete cosine transform to weaken the interference of background, and this segmentation was used as the initial contour of GCBAC to avoid artificial contour initialization. Then, dilation direction of contour line on both sides was changed to a one-way lateral dilation to avoid boundary shrinking bias. Results show that the proposed method can accurately locate pest region and clearly segment the contour of rice planthoppers.
